Robust feature matching using guided local outlier factor. (September 2021)

• Record Type: Journal Article
• Authors: Wang, Gang; Chen, Yufei
• Is Part Of: Pattern recognition. Volume 117(2021)
• Publication Date: 2021-09
• Subjects: Feature matching -- Mismatch removal -- Rejecting outliers -- Locality preserving -- Image matching; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.107986
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We propose a novel non-iterative approach for robust feature matching. Heavy outliers can be detected and removed by the guided local outlier factor. Multi-granularity neighborhood structure-preserving prevents the matching collapse. Abstract: Matching local features on two or more images is fundamental for many applications in the field of computer vision and pattern recognition. Identifying and rejecting mismatches is an important part in the framework of feature matching, due to the putative correspondences always contaminated by mismatches with the error-prone local feature detectors. In this paper, we introduce a novel method, namely Guided Local Outlier Factor (GLOF) for feature matching with gross mismatches under multi-granularity neighborhood structure-preserving. We first construct a tentative correspondence set by matching multi-features. Then, we identify and remove mismatches. Inspired by the anomaly detection technique, putative correspondences are assigned to a particular score, so abnormal instances, i.e., mismatches can be classified by a user-defined threshold. More specially, the neighborhood preserving guides the local searching procedure. Moreover, to eliminate the fluctuation of the matching results with different sizes of local neighbors, we use the multi-granularity algorithm to average out the deviation. Experimental results demonstrate that the introduced approach is superior to several state-of-the-art methods in terms of mismatch rejection on publicly available datasets.